Machine learning in trading: theory, models, practice and algo-trading - page 2481

New comment
 

Sometimes I still think that without proper statistics it's easy to write everything off as "wrong initial data" (although in fact they are what they usually are, there won't be any others)...

Here with the help of t-statistics man has examined the Flat/Trend of DAX ( https://dep_msm.pnzgu.ru/files/dep_msm.pnzgu.ru/metody_postroeniya_i_analiza_statisticheskih_modeley_vremennyh_ryadov.pdf )

By analyzing the autocorrelation function and correlogram, the structure of the series can be revealed.
If the highest was the autocorrelation coefficient of the first order, the studied series contains only a trend. If the highest autocorrelation coefficient is of order τ, the series contains cyclic fluctuations with periodicity in τ time moments

although, of course, everything is as usual history... if one believes it will repeat itself...

It is more correct to say that "it is not the previous prices that determine the current/future price", but OTHER factors

 
Dmytryi Nazarchuk #:

1. All mathematical methods for nonstationary processes are shamanism. Because you can only predict the future based on the past, and if the future does not depend on the past, predictions based on the past do not work.

Therefore, the choice of method, model, etc. does not play any role - only the right choice of input variables.

You can go no further

I agree...

What are your thoughts on the signs?

JeeyCi #:

sometimes I still think that without the right statistics it's easy to write everything off as "wrong nach data" (although in fact they are what they usually are, there won't be any others)...

Here with the help of t-statistics man has considered Flat/Trend of DAX ( https://dep_msm.pnzgu.ru/files/dep_msm.pnzgu.ru/metody_postroeniya_i_analiza_statisticheskih_modeley_vremennyh_ryadov.pdf )

Although, of course, it's all history as usual... if you believe it will happen again...

It is possible to make a simple correlation indicator, for example Mann-Kendall Trend Test with all the disadvantages of the usual indicator

There is no flat in the market, it's a subjective fiction.

 
mytarmailS #:

I agree.

What are your thoughts on the signs?

Try a cross-market analysis.

I've written to Bart Simpson that oil producer stock prices are directly influenced by oil prices, but there's a clinic with a sticking question mark in the keyboard...

 
mytarmailS #:
Only there is no flat in the market, it's a subjective fiction...

there is a Correction - which is achieved by OR depth, OR duration... The second is an abstraction called a "flatulence" (if you like)

 

By the way, the only point that remains to be determined is whether the indicator lags or outpaces... This fact should not always be considered unambiguously-constant-always-more-logical... It depends on the indicator (especially a self-written one and not tied to a time series)...

(provided that the indicators are trend forming ones and taking into account that the market usually slows down before finding a new balance - to the question of volatility)

 
Dmytryi Nazarchuk #:

Bullshit

Guess how it turns out, huh?

 
Dmytryi Nazarchuk #:

Try intermarket analysis.

I wrote to Bart Simpson that stock prices of oil companies are directly influenced by oil prices, but there's a clinic with a stuck question mark in the keyboard...

Well let's start with the fact that the opioid of the entire market is the national currency or deposit currency exchange rate.

And as luck would have it on Moex there is an instrument Si that satisfies our condition.

1. it is as vaolatile as the pound dollar intraday.

2. There is information on it on the open institute, as well as on other instruments.

3. the opener is a tax agent, so be warned!

For the rest, I take only 13 instruments from the market, by the way, maybe the number is not correct, but these instruments were chosen according to my instincts. Here is their list!

Name_instr[0]="Si Splice";
Name_instr[1]="BR Splice";
Name_instr[2]="LKOH Splice";
Name_instr[3]="ROSN Splice";
Name_instr[4]="RTS Splice";
Name_instr[5]="SBRF Splice";
Name_instr[6]="ED Splice";
Name_instr[7]="Eu Splice";
Name_instr[8]="GAZR Splice";
Name_instr[9]="MGNT Splice";
Name_instr[10]="VTBR Splice";
Name_instr[11]="MXI Splice";
Name_instr[12]="MIX Splice";
Name_instr[13]="GOLD Splice";

From this list, I form a training file of 7500 columns and 50 rows, where the first 10 rows I have a test, the rest in the work.

After pre-processing I have from 150 to 300 important columns for the target, and on this trained network by the method of reference vectors (one of the costly for machine learning) its complexity is that when you add 1 column to the model the complexity of the resulting polynomial is doubled, so I can not build a model of more than 15 inputs even on a super powerful computer. Well mine pulls 12, and the Mail's server does it no faster. This kind of raises the question of deeper meaning! There is a cost of getting a model, that is, every optimization it is not free and there is the potential to earn this model or get a loss. It does not always take much time to get a good model. You can get a good model at 5-7 inputs and at 10-12 and it imposes even more troublesome that if you can't get good training figures at 5-7 inputs, then you have to start training at 12, where up to 9 there is only a decrease and only then it may increase. In general, the process of optimization, I tell you. But I got accustomed to 2-4 hours of good work, get my models and smoke bamboo again, sit and watch. There is no other way :-(.

So what tools will be selected from the list above, I do not know, but at the expense of oil to say that drops it often, so I included it.

Dimitry, did I answer your clinical question?

 
That is, we need to think wider in the market, or rather to be one step higher. After all, the currency gives exactly the effect that everything else within the country becomes a step lower, companies, raw materials, resources. And with the help of them we try to forecast. Well, in any case, I, what and ALL invite to do exactly that. Let's put the heat on the volatility of Monday's opening on MOEX on Si ????????!!!!!!! I also have such a bad habit to leave positions after weekend, robot kuli :-(
 
Mihail Marchukajtes #:

(booze is drunk))

 
Mihail Marchukajtes #:
So you need to think more broadly in the market, or rather to be one step higher. After all, the currency gives exactly such an effect that everything else within the country becomes one step lower, companies, raw materials, resources. And with the help of them we try to make forecasts. Well, in any case, I, what and ALL invite to do exactly that. Let's put the heat on the volatility of Monday's opening on MOEX by Si ????????!!!!!!! I also have such a bad habit to leave positions after weekend, robot.
There are only gaps, it's a waste of money
1...247424752476247724782479248024812482248324842485248624872488...3496
New comment
Reason: